Emotional development in preschool children is a core aspect of psychological development; however, their emotional expressions are characterized by intuitiveness, instantaneity, and instability, posing challenges for traditional support models that struggle with balancing comprehensive and non-intrusive data collection and suffer from homogenized intervention strategies. This paper focuses on the innovation of AI-empowered support models for preschool children’s emotional development, analyzing their technological logic and feasibility. By leveraging technologies such as the Internet of Things, computer vision, and speech recognition, it achieves precise and unobtrusive collection and systematic processing of multidimensional emotional data. Relying on machine learning and affective computing, personalized emotional recognition models tailored to different age groups and scenarios are constructed, leading to the generation of differentiated intervention strategies suitable for both kindergarten and home settings. On this basis, an overall framework of technology-supported, dual-field collaborative, and closed-loop operation is established, innovatively proposing three sub-models: precise intervention in kindergartens, intelligent co-education in families, and seamless data integration between home and kindergarten. To ensure the implementation of these models, an AI-assisted, human-led human-machine collaboration mechanism, a dynamic assessment and feedback mechanism, as well as an ethical framework and privacy security system encompassing principles of informed consent, minimal necessity, and non-maleficence are established.